Parallelization of Shallow-water Equations with the Algorithmic Skeleton Library SkelGIS

Giving an easy access to non computer-scientists to high performance computing is a very important challenge of the research in computer sciences. SkelGIS is an algorithmic skeleton library that aims at writing efficient parallel programs in a sequential way. In this paper, a parallel implementation of shallow-water equations using SkelGIS is presented. This implementation is compared, in terms of learning efforts and performances, to an MPI (Message Passing Interface) implementation. New concepts of algorithmic skeletons are proposed by SkelGIS, and offer a better expressiveness to the user than existing skeleton solutions. As a result, SkelGIS enables an efficient parallelization of complex numerical method with algorithmic skeletons, and therefore without any knowledge on high performance computing.